Intelligent Peer-To-Peer Text Data Augmentation

This application claims the benefit of U.S. Patent Application No. 63/654,451, filed on May 31, 2024, and to International Application No. PCT/IB2024/056820, filed on Jul. 12, 2024, the contents of which are incorporated herein by reference in their entirety.

Peer-to-peer messaging platforms have become a popular means for modern communication, allowing individuals to connect over the internet. However, as popular as these platforms have become, they encounter the problem of data opacity. These systems oftentimes act like black-boxes and leave users unaware of how their conversation data is being processed behind the scenes. This lack of transparency raises concerns about data privacy and data security.

Furthermore, existing methods struggle to intelligently integrate relevant contextual information such as advertisements into conversations without compromising data security. Peer-to-peer messaging platforms do not have a reliable channel for delivering personalized and contextually relevant content to their users within their conversations. As a result, advertisers resort to opaque methods that may rely on sensitive conversation data to push advertisements on other platforms. These trends only increase data opacity surrounding the possibility of private conversations being analyzed and exploited for targeted advertising or other purposes without consent or awareness.

In the drawings, like reference numbers generally indicate identical or similar elements. Additionally, generally, the left-most digit(s) of a reference number identifies the drawing in which the reference number first appears.

Disclosed herein are system, apparatus, device, method, and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for augmenting conversation text data to reduce data opacity and improve data transparency of advertising methods within peer-to-peer messaging platforms.

In today's digital age, peer-to-peer messaging platforms allow individuals to communicate and share information seamlessly over the Internet, many of which promise data security and privacy to their users. However, these platforms usually operate in a black-box fashion, leaving users unsure whether their data is truly secure and whether their data is being processed behind the scenes.

A common phenomenon occurs where users may encounter advertisements on different platforms or websites for products or services they have discussed in supposedly private conversations. This phenomenon raises concerns about data privacy and potential misuse of sensitive data for targeted advertising purposes. And while the targeted advertisements might just be a result of unconscious user interactions with content the users are interested in, the lack of transparency surrounding the use of conversation data for advertising purposes still remains a pressing issue. Users should be allowed to know how their data is being managed within peer-to-peer messaging platforms. As such, a core technical issue lies in the data opacity of current methods for intelligently processing conversation text data in peer-to-peer messaging platforms for generating personalized contextual information, such as advertisements.

Various embodiments in accordance with the present disclosure overcome the aforementioned technological problems by intelligently augmenting conversation text data in a peer-to-peer messaging platform. A text augmentation system may first receive raw conversation text data from a peer-to-peer messaging platform. The text augmentation system may then leverage a natural language processing (NLP) service to identify any keywords in the raw conversation text data that may be associated with a desired product or service. For example, a text augmentation system may receive a message that reads: “Doing good as well. Planning a vacation. Can you recommend a good hotel in Dubai?” The NLP service may analyze the raw text data, identify the keywords “hotel in Dubai”, and determine that the user is interested in booking a hotel in Dubai. The text augmentation system may then query a data store using these keywords for any existing associated advertisements. If multiple advertisements are found, text augmentation system may then select one or more of the advertisements using an advertisement auction service. Upon selecting the one or more advertisements, the text augmentation system may then transmit the identified keyword and associated advertisement or advertisements to the peer-to-peer messaging platform to augment the conversation text data.

In some embodiments, the conversation text data may be augmented by embedding a hyperlink within the identified keywords. In some embodiments, the conversation text data may be augmented by adding a popup or dropdown next to the identified keywords. In some embodiments, the conversation text data may be by any means that may differentiate the keyword text from the rest of the message. For example, the keyword text may be highlighted. The keyword text may also be given a different font color from the other words in the message. The keyword text may also be given a shimmer or shining text effect. By doing so, users may see exactly which words in a message are being processed for generating relevant advertisements. This allows them to accurately trace the flow of data, thereby removing data opacity concerns surrounding how subsequent advertisements are generated, especially ones outside of the peer-to-peer messaging platform.

In addition to reducing data opacity and improving data transparency, this text augmentation system has numerous additional technological and practical benefits. First, by leveraging modern NLP services, the text augmentation system more accurately determines which keywords in a message are actually associated with a product or service that a user can benefit from. In some embodiments, the NLP service may extract additional contextual information from the message or previously sent messages to help determine which keywords should be identified. In a non-limiting example, a message might read: “I am so torn between buying a cat and buying a dog. Dad would be so upset if I got a dog though, so maybe I'll just get a cat.” Given this message, the NLP service may determine that the user wants both a “cat” and a “dog” but ultimately only identify “cat” as the keyword, since the user has expressed concerns about buying the dog. By leveraging modern NLP services, the text augmentation system is able to intelligently identify the keyword to be processed. As a result, users may be more inclined to accept and interact with these more personalized advertisements.

Second, augmenting the conversation text data is non-intrusive and does not worsen the user experience, whereas conventional advertising methods are often intrusive and common pain points for users. These may include, but are not limited to, display ads that appear as banner or static advertisements, pop-up ads that automatically open new windows or overlays, and auto-redirect ads, which automatically redirect users to different sites oftentimes for click fraud. By simply augmenting the conversation text data, the advertisement is not forced upon the user, and the user has full control to decide whether or not they want to explore the associated advertisement. In fact, users may end up being more inclined to accept and interact with these advertisements due to their non-intrusive and subtle nature.

Third, augmenting the text data within the conversations themselves streamlines the product or service searching process for users. In a non-limiting example, a user might already be planning on setting aside some time to look for a specific product or service online. By augmenting the conversation text data with an advertisement, the user may skip the searching process altogether and connect to a vendor directly. In some embodiments, the system may be expanded to incentivize users to leverage the vendors within the embedded advertisements over other options by providing certain discounts or features when accessing the vendors through the messaging platform's augmented conversation text data. In a non-limiting example, the messaging platform may augment the conversation text data with a special link that applies a discount code on the next transaction made on the vendor website. As such, augmenting the conversation text data provides the user with seamless and streamlined access to products and services they are interested in.

Fourth, augmenting the text data within a conversation may facilitate additional functionalities that may be useful for a user. In some embodiments, the augmented text data may be configured to provide contextual information within a conversation. For example, a message may read, “I wonder how high Mt. Everest is.” In response, the augmented text data may present the requested information. In some embodiments, the augmented text data may be configured to integrate third party services into the messaging application. For example, rather than or in addition to an advertisement, an augmented conversation text data may initiate an embedded taxi booking service within the messaging application. An integrated embedded application may save the user a trip to the advertiser's website to perform a desired task.

is a block diagram of a systemillustrating example functionality for a text augmentation system (TAS), according to some embodiments. The example systemis provided for the purpose of illustration only and does not limit the disclosed embodiments. TASmay augment conversation text data to reduce data opacity and improve data transparency. Example systemmay include TAS, user device, messaging platform, and marketplace. TASmay include data processing service, auction service, natural language processing (NLP) service, TAS gateway, and data store. Marketplacemay include marketplace gateway.

In some embodiments, TASmay be a software development kit (SDK) installed directly on messaging platform. In some embodiments, TASmay operate as a cloud service that messaging platformcan subscribe to. TASmay receive raw conversation text data from messaging platform, which is entered by user device. For example, TASmay receive a message that reads: “Doing good as well. Planning a vacation. Can you recommend a good hotel in Dubai?” Upon receiving the raw conversation text data, TASmay employ data processing serviceto analyze and process the raw conversation text data. In some embodiments, data processing servicemay analyze the raw conversation text data to identify a keyword or set of keywords that are associated with a desired product or service by a user. Messaging platformmay be any be any messaging application service, communication platform, or application with a chat functionality (i.e. a chatbot).

In some embodiments, data processing servicemay employ NLP serviceto identify the keyword or set of keywords. NLP servicemay be a large language model or artificial intelligence proxy that connects with closed source enterprise application program interfaces (APIs). In some embodiments, NLP servicemay be an open source model. These powerful models are able to analyze conversation text data and accurately identify where and when a user might benefit most from an advertisement. Any known or future large language models may be substituted without departing from the scope of the technology described herein.

A key difference between utilizing these powerful large language models over traditional NLP or rule-based methods is that large language models are able to understand and extract additional semantic information from conversation text data such as context and emotions. In a non-limiting example, NLP servicemay receive a message that reads: “I really want to buy a new Mercedes, but I have only a few thousand dollars in my bank account. Instead, I will buy a used Toyota.” Traditional NLP or rule-based methods may identify that the user wants to either buy a Mercedes, a used Mercedes, a Toyota, or a used Toyota. However, NLP servicemay extract the needed contextual information from the rest of the message, such as the user not having enough money to buy a new Mercedes. As a result, NLP servicemay just identify the keyword “used Toyota” based on the conversation text data and the extracted context. This result may more accurately align with the user's actual intentions and needs.

Upon identifying the keyword or keywords, data processing servicemay query data storefor one or more advertisements associated with the identified keyword. Each advertisement in data storemay be associated with a corresponding advertisement campaign. As used herein, an advertisement campaign may refer to an advertisement and any associated metadata including but not limited to advertisement budget, advertisement bidding values, advertisement location targeting, effective time windows, and advertisement statistics. In some embodiments, data storemay contain keyword-advertisement pairings configured by advertisers through marketplace.

In some embodiments, the query initiated by data processing servicemay return an exact keyword match. For example, if data processing servicequeries data storewith the keywords “cheap Lakers tickets”, the query may only return advertisements with the exact keywords “cheap Lakers tickets.” In some embodiments, the query initiated by data processing servicemay return any similar or associated keyword-advertisement pairings. For example, a query using the same keywords from before may return advertisements with keywords such as “Lakers tickets”, “Lakers tickets on sale”, “best Lakers tickets”, “Lakers game”, and more. In some embodiments, data processing service may employ vector or string similarity algorithms to determine a similarity score between the queried keywords and the keywords inside data store. In some embodiments, TASmay define a preconfigured threshold value to determine which keyword-advertisement pairings are returned, based on calculated similarity scores falling within the threshold value. In some embodiments, a threshold value may be configured by the advertisers, the messaging platform, or the user for additional customization and/or personalization.

In some embodiments data processing servicemay employ auction serviceto select an advertisement or advertisements from among the results returned by the query. In some embodiments, auction servicemay include a proxy connected to a third-party API. When making a selection, auction servicemay analyze values from the advertisements' corresponding advertisement campaigns. In a non-limiting example, auction servicemay analyze advertisement bid, advertisement quality, advertisement location target, effective time windows when conducting the auction. In some embodiments, auction servicemay calculate an auction score for each query result by weighting each metadata value and summing the weighted values together. Auction servicemay select the advertisement with the highest auction score to be returned to data processing service.

In some embodiments, an auction score can be determined using a machine learning model. The machine learning model may be trained to accept advertisement campaign metadata values and make a score prediction. The machine learning model may be a classification or regression model, including but not limited to Naïve Bayes, Support Vector Machines, Decision Trees, Random Forest, Hidden Markov Models, Neural Networks, Gradient Boosting Machines, and Gaussian Mixture Models. In a non-limiting example, the training data may include labeled auction scores that reflect real world advertisement performance. For example, advertisements that led to successful user interaction and completed transactions may receive a high auction score. Similarly, advertisements that did not lead to a successful user interaction or completed transaction may receive a low auction score. In some embodiments, advertisements with the highest running total transaction amounts may receive a high auction score. Similarly, advertisements with low running total transaction amounts may receive a low auction score.

A machine learning model may better capture the relationships between the different advertisement campaign metadata values compared to a blind weighting strategy. This may be the case especially when the data is non-linear in nature. By training on real world data, the machine learning model may learn certain trends. In a non-limiting example, the machine learning model may learn that the campaigns with the highest bids do not always necessarily create the most successful advertisements. In another non-limiting example, the machine learning model may learn that campaigns with high advertisement quality but insufficient bid values may not survive the auction process and appear in front of users enough times to generate clicks and revenue.

In some embodiments, advertisement campaigns may be configured by advertisers in marketplace. Marketplacemay provide an accessible and interactive interface for advertisers to initiate and manage different advertisement campaigns. Marketplacemay be a mobile application or a web application. TASmay communicate with marketplacethrough secure API gateway communication channels TAS gatewayand marketplace gateway. TASmay store the configured advertisement campaign metadata inside data store.

In some embodiments, advertisers may define their desired keywords and build corresponding advertisements to those keywords. In some embodiments, advertisers may define bidding values and strategies associated with each advertisement that may be analyzed by auction servicewhen conducting auctions. In some embodiments, advertisers may define a maximum bid or a minimum bid. Advertisers may also define a budget associated with each advertisement and a corresponding bidding strategy including but not limited to cost per click (CPC), cost per thousand impressions (CPM), and cost per acquisition (CPA). A CPC strategy may involve setting bids to maximize the number of times an advertisement may be clicked within a budget. A CPM strategy may involve setting bids to maximize the number of impressions or views an advertisement may receive within a budget. A CPA strategy may involve setting bids to maximize the number times a new customer may be acquired within a budget.

In some embodiments, marketplacemay display different statistics associated with an advertisement campaign. In a non-limiting example, marketplacemay display the running total of clicks, impressions, click-through rate (CTR), and billing fees for each advertisement campaign during its lifetime. CTR may refer to the rate at which an advertisement was clicked after being viewed. In some embodiments, auction servicemay utilize the marketplace statistics when conducting auctions between various advertisement campaigns. In a non-limiting example, an advertisement with a higher CTR may receive a higher weighting when calculating a corresponding auction score. In another non-limiting example, the CTR may be included as a feature when training a machine learning model for predicting auction scores and making inferences with the machine learning model.

In some embodiments, TASmay transmit the associated advertisement and keywords back to messaging platformafter receiving the results of the auction. Messaging platformmay then augment the conversation text data using the received advertisement or advertisements and associated keywords. In some embodiments, messaging platformmay embed the selected advertisement as a hyperlink within the identified keyword. The advertisement may include an advertisement title, advertisement image, and advertisement Uniform Resource Locator (URL). In some embodiments, a direct hyperlink to the advertisement URL may be embedded within the identified keyword. In some embodiments, the messaging platformmay embed the advertisement as a popup or dropdown element next to the identified keyword. In some embodiments, messaging platformmay augment the conversation text data by any means that may differentiate the keyword text from the rest of the message. In a non-limiting example, messaging platformmay highlight the background surrounding the keyword. Messaging platformmay also change the text color of the keyword. After augmenting the conversation text data, messaging platformmay display the augmented message to the user through user device. As a result of this message augmentation, the user may identify exactly which keyword is being associated with an advertisement.

is a block diagramillustrating an example marketplace, according to some embodiments. Marketplacemay be an example of marketplace(of) and may communicate with TASto facilitate the augmentation of conversation text data. The example marketplaceincludes various components, however it is understood that in other embodiments, marketplacemay include components in addition to or different from those described below.

In some embodiments, advertiserand administratormay interact with marketplace. In some embodiments, marketplacemay include frontend services, application core, client user interface (UI), client backend, shared entities service, redirection service, admin UI, admin backend, statistics service, and marketplace gateway. Marketplace gatewaymay be an example of marketplace gateway(of). Application coremay include campaign service, accounts service, and billing service.

In some embodiments, advertisermay interact with marketplacethrough client UI. Client UI may leverage frontend servicesand client backendto connect to application core. In some embodiments, advertisermay wish to initiate and manage campaigns with marketplace. Advertisermay first create an account to manage different campaigns on marketplace. Marketplacemay employ application coreto create an account using accounts service. Advertisermay also set up billing information through marketplace. In some embodiments, application coremay employ billing serviceto configure the billing information entered by advertiser. After advertisercreates an account and configures billing information, advertisermay begin building an advertisement campaign.

In some embodiments, advertisermay enter an advertisement and associated keywords through client UI. In some embodiments, an advertisement may include an advertisement title, advertisement photo, and advertisement URL. In some embodiments, advertisermay configure a keyword matching strategy and define how closely the conversation text data should match the keywords. In some embodiments, advertisermay define a phrase matching strategy that will include the advertisement in any message containing the same meaning as the keywords. For example, phrase matching using the keyword “Lakers tickets” may identify “LA basketball game” or “Laker event Saturday.” In some embodiments, advertisermay define a broad matching strategy that will include the advertisement in any message that relates to the keywords. For example, broad matching using the keyword “Lakers tickets” may identify “basketball tournament” or “AMC movie tickets.” In some embodiments, advertisermay define an exact matching strategy that will include the advertisement only in messages that contain the same keywords and meaning inside the message. For example, exact matching using the keyword “Lakers tickets” may identify “Lakers game” or “Lakers match”.

In some embodiments, advertisermay define bidding values and strategies associated with each advertisement. In some embodiments, advertisermay define a maximum bid or a minimum bid. Advertisermay also define a budget associated with each advertisement and a corresponding bidding strategy including but not limited to cost per click (CPC), cost per thousand impressions (CPM), and cost per action (CPA). A CPC strategy may involve setting bids to maximize the number of times an advertisement may be clicked within a budget. A CPM strategy may involve setting bids to maximize the number of impressions or views an advertisement may receive within a budget. A CPA strategy may involve setting bids to maximize the number times customer actions (e.g. sales, sign-ups, installs, etc.) may be performed within a budget.

In some embodiments, marketplacemay redirect users to an advertisement's URL using redirection servicewhen an advertisement is clicked on a user device. In some embodiments, marketplacemay employ shared entities serviceto manage any dynamic variables shared across the marketplace services. In some embodiments, marketplacemay include a single sign-on (SSO) service that authenticates advertiserand administratorinside marketplace. In some embodiments, administratormay monitor and manage marketplacethrough admin UIand admin backend. In some embodiments, marketplacemay display different statistics associated with an advertisement campaign using statistics service.

In a non-limiting example, statistics service may track the running total of clicks, impressions, click-through rate (CTR), and billing fees for each advertisement campaign during its lifetime. In some embodiments, statistics servicemay also calculate different fee related statistics in addition to running totals of billing fees such as but not limited to average cost-per-click (CPC), cost-per-view (CPV), and cost-per-action (CPA). CPC may refer to the average amount an advertiseris charged per clicked advertisement in the campaign. CPV may refer to the average amount an advertiseris charged per impression or view of an advertisement in the campaign. CPA may refer to the amount an advertiseris charged per successful customer action from the advertisement, such as but not limited to a product purchase, an installation, a sign-up, etc.

illustrates an example text augmentationA, according to some embodiments. Text augmentationA shall be described with reference to TAS(of). However, text augmentationA is not limited to that example system. The text augmentation provided inis merely exemplary, and one skilled in the relevant art(s) will appreciate that many approaches may be taken to provide a suitable text augmentationA in accordance with this disclosure. In some embodiments, text augmentationA may include user device, messaging application, augmented message, hyperlinked keyword, advertisement, advertisement image, and advertisement link.

In some embodiments, messaging platformmay augment conversation text data in messaging application. For example, messaging platformmay augment a conversation message that reads: “Doing good as well. Planning a vacation. Can you recommend a good hotel in Dubai?” Messaging platformmay then transmit the contents of the conversation message to TAS. TASmay then analyze the contents of the conversation message to identify a keyword associated with a desired product or service. For example, TASmay determine that a user wants to book a hotel in Dubai and identify the keyword “hotel in Dubai.” In some embodiments, TASmay identify multiple keywords inside the conversation message. For example, TASmay have also identified “vacation” as a keyword. TASmay then query a data store for one or more associated advertisements with keyword.

In some embodiments, TASmay receive multiple advertisements associated with the keyword as a result of the query. Then, TASmay select an advertisement or advertisements among the results using auction service. In some embodiments, auction servicemay generate auction scores for each result and return the result with the highest auction score. In some embodiments, auction servicemay return multiple results. For example, auction servicemay return the advertisements with the three highest auction scores. In some embodiments, auction servicemay return all of the advertisements with auction scores above a preconfigured threshold value. The auction score of an advertisement may be related to expected financial success, which may be measured by advertisement statistics such as CTR and advertisement clicks. For example, a higher CTR may imply higher user engagement, which may lead to a higher auction score. In another example, more advertisement clicks or impressions may also imply user engagement, which may also lead to a higher auction score.

In some embodiments, TASmay transmit the result of auction serviceback to messaging platform. Messaging platformmay then augment the conversation data to produce augmented message. In some embodiments, messaging platformmay produce augmented messageby embedding the returned advertisement as a hyperlink within the keyword and create hyperlinked keyword. In some embodiments, messaging platformmay produce augmented messageby embedding the returned advertisement as a popup or dropdown next to the keyword.

In some embodiments, a user may wish to view the contents of augmented messageby interacting with hyperlinked keyword. For example, a user may click on an underlined “hotel in Dubai” inside an augmented message that reads “Doing good as well. Planning a vacation. Can you recommend a good hotel in Dubai?” In some embodiments, advertisementmay be displayed on user device. Advertisementmay show advertisement details such as but not limited to advertisement title, advertisement description, advertisement URL, advertisement image, advertisement video, advertisement audio or other media content, and advertisement link. Advertisementmay also display an address, a telephone number, a mode of operation, prices, an application form, promotions, and additional links. In some embodiments, a user may wish to interact with advertisementto view the advertised product or service. For example, a user may click advertisement link. In some embodiments, advertisement linkmay redirect a user to an advertisement URL or website. In some embodiments, advertisement linkmay initiate an additional application associated with the advertisement embedded within the messaging application chat. For example, a user may click an advertisement linkassociated with a taxi service and be presented with an embedded application to book a taxi ride within the messaging application. In some embodiments, messaging applicationmay display multiple advertisements returned by the auction. In a non-limiting example, a user may be able to swipe or click between the different advertisements displayed in messaging application. As a result, the user may have multiple reference points to compare an initial advertisement. In some embodiments, the advertisement linkmay be a special discount link that provides a user with a discount when purchasing a product or service. This may serve to incentivize users to participate in the augmented messaging environment and interact with the advertisements.

illustrates another example text augmentationB, according to some embodiments. This example includes similar components as, namely user device, messaging application, augmented message, and hyperlinked keyword. However, instead of advertisement, text augmentationB includes embedded taxi service application, pick-up location, drop-off location, and book taxi button. In some embodiments, text augmentationB may be displayed as a direct result from a user interaction with augmented conversation text data, such as a user clicking a hyperlinked keyword. In some embodiments, text augmentationB may be displayed after a user clicks an advertisement link, such as advertisement link(of). In some embodiments, embedded applicationmay be connected to a third-party service. For example, a user may connect to a taxi service through embedded applicationby entering a pick-up locationand drop-off locationand clicking book taxi button. As a result, the embedded application may find and book a relevant taxi for the user automatically.

illustrates an example methodof augmenting conversation text data, according to some embodiments of the disclosure. As a convenience and not a limitation,may be described with regard to elements of. The example methodmay represent the operation of devices (e.g. messaging platform, data processing service, auction service, marketplace gateway, data store, or a combination thereof of) implementing augmenting conversation text data. The example methodmay also be performed by computer systemof. But the example methodis not limited to the specific embodiments depicted in those figures and other systems may be used to perform the method, as will be understood by those skilled in the art. It is to be appreciated that not all operations may be needed, and the operations may not be performed in the same order as shown in.

At, messaging platformmay transmit a message to be augmented to data processing serviceof TAS. In some embodiments, messaging platformmay also send previous messages in the conversation history or the entire conversation history to data processing servicealong with the message to be augmented. For example, messaging platformmay want to provide context for the message to be augmented.

At, data processing servicemay perform a keyword search to identify a keyword from the message to be augmented that is associated with a desired product or service by a user. In some embodiments, data processing servicemay employ NLP serviceto intelligently extract the keyword from the message. In some embodiments, data processing servicemay also include the previous messages or entire conversation history associated with the message to be augmented when invoking NLP serviceto provide additional context and produce a more accurate result. For example, data processing servicemay receive an ambiguous message that reads, “Bucks would be a good choice.” If data processing serviceonly receives this message and no other context, NLP servicemay not know what the message is about. For example, “Bucks” could have multiple meanings, many of which are not common items people would look to purchase (e.g. the basketball team, male deer, money, etc.). Therefore, to be statistically safe, NLP servicemay not identify a keyword. However, data processing servicemay receive previous conversation history that reveals that the user is looking to buy some shoes for an upcoming backpacking trip. As a result of this added context, NLP servicemay understand that “Bucks” in the message refers to a type of shoe and thus may return the keyword “Bucks”. Having additional context will help the NLP servicereturn accurate keywords that actually reflect the intentions of the user.

At, data processing servicemay query data storewith the results of the keyword search for associated advertisements and campaign metadata. In some embodiments, the query may employ vector or string similarity algorithms to determine a similarity score between the queried keywords and the keywords inside data store. In some embodiments, advertisement campaigns may define a keyword matching strategy for the query to consider. For example, an advertisement campaign may define a phrase matching strategy which may return the keyword in queries for items containing the same meaning. In another example, an advertisement campaign may define a broad matching strategy which may return the keyword in queries for related items. In another example, an advertisement campaign may define an exact matching strategy, which may return a keyword in queries for items with the same meaning.

At, data storemay return the query results to data processing service. In some embodiments, data storemay return keyword-advertisement pairings as a result of the query. In some embodiments, campaign metadata may include advertisement budget, advertising bidding values, advertisement location targeting, effective time windows, and advertisement statistics.

At, data processing servicemay send an auction request to auction serviceto trim the results of the keyword search. The auction request may include the keywords returned by the query and the campaign metadata. In some embodiments, data processing servicemay employ a local system to execute the auction request. In some embodiments, data processing servicemay invoke a third party API to perform the auction.

At, auction servicemay conduct an auction to select and return an advertisement from the auction request. In some embodiments, auction servicemay calculate an auction score for each query result by weighting each associated campaign metadata value and summing the weighted values together. In some embodiments, the auction score may be determined by a machine learning model trained to accept campaign metadata values and make a score prediction. The machine learning model may be trained on real world advertisement performance, where successful advertisements may be given higher auction scores, and less successful advertisements may be given lower auction scores. In some embodiments, the auction scores may be influenced by advertisement statistics. For example, an advertisement with a higher CTR may be given a higher auction score. In another example, a newly created advertisement with few clicks and impressions may be given a lower score.

At, auction servicemay return the results of the auction to data processing service. In some embodiments, auction servicemay return multiple auction results. In a non-limiting example, auction service may be configured with a threshold auction score value, where any advertisements with scores above the threshold may be returned.

At, data processing servicemay augment the message with the results of the auction. In some embodiments, data processing servicemay embed an advertisement inside an associated keyword of conversation text data. In some embodiments, data processing servicemay embed an advertisement inside a popup or dropdown next to the associated keyword of conversation text data. Alternatively, data processing servicemay choose not to augment the message.

At, data processing servicemay transmit the augmented message to messaging platform. Alternatively, data processing servicemay transmit the raw keyword-advertisement pairings returned by the auction to the messaging platform. In this case, the messaging platform may augment the conversation text data accordingly.

At, messaging platformmay display the augmented message to the user. In some embodiments, messaging platformmay re-render conversation text data to show the augmented message. For example, the conversation text data may be re-rendered to underline the keyword associated with the advertisement. This way, the user may clearly see which word(s) and message(s) are associated with the advertisement.